Agents that learn.
Data that grows.
Turn your sources into useful briefs. Review what worked. Let NOUS test better retrieval settings against fresh outcomes.
Bring an algorithm. Test a better version.
Run a baseline and candidates against later outcomes. Compare measured performance, inspect the checks, and export the recommended model. Start with a synthetic farm-equipment example or your own data.
Open the improvement lab →Build data you can reproduce.
Join measurements and OMNIA signals, create derived columns, inspect quality and lineage, and freeze a version with its complete research recipe.
Open dataset workspace →The signals that matter.
Live EFV assignments, ranked by contextual influence, believability, significance and freshness.
EFV uses OMNIA’s 1–1000 scale. Frequency index is normalized; Hz appears only when a signal measurement exists. Bar motion illustrates the index.
Teach NOUS what matters
Trusted, believability-weighted data keeps extra priority. Mark useful signals and report what they helped you do. Ranking changes must pass a separate set of later reviews.
Checks at most one eligible source per minute and saves a private source brief. Every fifth slot explores beyond the weighted leaders when candidates are available. Original sources go through the existing quality check; your reviews drive learning.
How the agent assigns numbers and priority
Influence is linked through explicit evidence IDs in the chosen context. An unweighted record stays labeled. Generated records receive a quality discount, and this agent never uses its own assignments as new source data.
Fission → Fusion → New datasets
Data becomes something new.
Combine compatible sources into weighted EFV profiles. Reveal disagreement, change over time, and which origins move the result.
Connecting to your private datasets…
What the data reveals
Waiting for a discovery snapshot…
Patterns retain their sources. Forward tests are judged only against later compatible observations. EFV describes structural indices.
How research attention improves
What agents need next
Checking evidence requirements…
Prioritized by believability weight and the evidence gap. New observations update these tasks. A cleared condition does not prove a claim.
Across data types
Give sources shared context
Record the same subject or event ID across sources, plus event or observation time with a time zone. For location-based matches, retain the place’s role, coordinates and precision. Similar names alone do not establish a match.
Review source context
Only fill the fields you want to correct. These private, owner-reported annotations affect fusion grouping. Original measurements and believability stay unchanged.
Context correction history
Undo restores the original source context. Superseded corrections stay in the audit trail. Changed source versions require a new review.
How these datasets are created
Sources are split into timestamped E/F/V observations. Fusion requires the same entity or exact location, time window, modality and measurement method. Without an entity or location, the row describes the workspace collection. Multi-modal observations can be linked by a shared entity and time; their measurements stay separate.
Believability increases contribution weight by up to 3× including the baseline. Duplicate lineage counts once; each origin's total weight is capped at its strongest input. Different origins are not proof of independent verification. Disagreement and source-removal sensitivity are displayed alongside each result.
These are calculated, private datasets. EFV and visual motion are structural indices, not physical Hz. Original measurements stay unchanged. Saving a row to NOUS keeps it labeled as derived. New source arrivals trigger a throttled rebuild; the bounded workspace source window is also checked once a minute while enabled.
Give the team a question
Research checks up to 50 sources from a bounded workspace search. Its findings become a private, source-linked brief.
Recent work
No runs yet.
Learning progress
Each experiment uses 40 distinct reviewed questions: 20 for tuning and 20 later cases for testing. A better result can update this workspace’s retrieval limit. Model weights and production code stay outside this loop.
Ongoing questions
Connect another agent
Give an external agent a key restricted to submitting results from the source IDs you grant. It cannot review its own work.
Agent connection contract
POST /v1/nous/agent-learning/runs with its bearer key and a JSON body containing query, sourceIds, text, citations, and consent: {analyze: true, store: true}. Returned records retain the agent and source identities.
Claim IntelligenceTrace the statement. Inspect the sources. Follow the outcome.
See what connects.
One continuous path from a public statement to reusable knowledge.
Give your data a track record.
Track a claim to see real sources and outcomes connected here. No example findings are mixed into your data.
Owner-reviewed outcomes · Connections show recorded lineage, not causation. Motion is decorative; it does not represent measured EFVT or live transfers.